Covalent inhibition of pro-apoptotic BAX

Abstract

BCL-2-associated X protein (BAX) is a promising therapeutic target for activating or restraining apoptosis in diseases of pathologic cell survival or cell death, respectively. In response to cellular stress, BAX transforms from a quiescent cytosolic monomer into a toxic oligomer that permeabilizes the mitochondria, releasing key apoptogenic factors. The mitochondrial lipid trans-2-hexadecenal (t-2-hex) sensitizes BAX activation by covalent derivatization of cysteine 126 (C126). In this study, we performed a disulfide tethering screen to discover C126-reactive molecules that modulate BAX activity. We identified covalent BAX inhibitor 1 (CBI1) as a compound that selectively derivatizes BAX at C126 and inhibits BAX activation by triggering ligands or point mutagenesis. Biochemical and structural analyses revealed that CBI1 can inhibit BAX by a dual mechanism of action: conformational constraint and competitive blockade of lipidation. These data inform a pharmacologic strategy for suppressing apoptosis in diseases of unwanted cell death by covalent targeting of BAX C126.

BCL-2-associated X protein (BAX) is a pro-apoptotic BCL-2 family protein that resides in the cytosol as a latent monomer until triggered to undergo a conformational change, which drives its mitochondrial translocation, homo-oligomerization and permeabilization of the mitochondrial outer membrane, resulting in programmed cell death¹. The structure of BAX consists of nine α-helices, eight of which surround a core hydrophobic α5 helix². For such a small protein of 21 kDa, monomeric BAX has multiple functional binding interfaces. At its N-terminal surface, α1 and α6 form a trigger site that initiates the conformational activation of BAX upon engagement by the BCL-2 homology 3 (BH3) α-helices of select pro-apoptotic family members, such as BID and BIM³. The first structural change involves displacement of the α1–α2 loop that overlies the trigger site and consequent exposure of the BAX BH3 helix (α2), which, in turn, can trigger other monomers of BAX to undergo conformational activation⁴. BH3 interaction at the trigger site also causes an allosteric change that is transmitted through the α5 core of the protein to a groove at the C-terminal surface⁴. Whereas the α9 helix of BAX is constitutively bound to this C-terminal groove, BH3 triggering causes allosteric release of α9, which then targets BAX to the mitochondrial outer membrane. Upon self-association into a yet unknown higher-order structure, BAX oligomers disrupt the integrity of the mitochondrial outer membrane via discrete subcomponents, such as its α6 helix, which has membrane-lytic properties⁵.

Importantly, the lethal consequence of renegade BAX activation requires that its monomeric structure be stabilized from within by interactions between key residues of the core α5 helix and those of the surrounding α-helices. Point mutagenesis of select residues of α5 destabilizes the BAX structure, lowering the threshold for both auto-activation and BH3-triggered activation⁶. Indeed, physiologic activation of BAX relies on destabilization of the α5 core, whether by direct BH3 triggering or the sensitizing effect of the mitochondrial lipid trans-2-hexadecenal (t-2-hex), a product of sphingolipid metabolism⁷. We recently determined that t-2-hex influences the conformational activation of BAX by non-enzymatic covalent derivatization of cysteine 126 (C126), which is located at the exposed junction of the α5/α6 hairpin⁸. Nuclear magnetic resonance (NMR) analysis of ¹⁵N-BAX upon t-2-hex titration revealed chemical shift perturbations in the core α5/α6 hairpin, in addition to allosteric effects involving the α1–α2 loop, α2 and α9, the very regions implicated in the conformational activation of BAX⁸. These data suggest that small molecules capable of selective covalent targeting of C126 could affect the stability of the BAX core and, thereby, modulate BAX function for potential therapeutic benefit.

Disulfide tethering screens have emerged as a powerful strategy for small-molecule discovery⁹. In this method, molecular fragments bearing disulfide warheads non-covalently engage compatible binding surfaces and then react with a native or installed cysteine in the region by thiol-disulfide exchange, with added reducing agents such as β-mercaptoethanol (BME) increasing the stringency of selection¹⁰. This screening strategy was successfully applied, for example, to identify small-molecule inhibitors that target C12 in mutant K-Ras (G12C)¹¹ and C55 in anti-apoptotic BFL-1 (ref. 12). Here, we applied disulfide tethering to identify small-molecule prototypes that covalently target C126 of BAX and interrogate both the mechanistic and functional consequences. In contrast to t-2-hex that sensitizes BAX activation upon covalent derivatization of C126, we discovered a small molecule that likewise targets C126 yet inhibits the conformational activation of BAX, informing the development of covalent BAX inhibitors as next-generation cytoprotective agents.

Results

Disulfide tethering identifies a covalent inhibitor of BAX

To identify small molecules capable of covalent targeting of BAX in a manner analogous to post-translational modification of C126 by t-2-hex (Fig. 1a), we screened a library of 1,600 disulfide-containing compounds^13,14 for reactivity with wild-type BAX in the presence of 500 μM BME for 1 h at room temperature, followed by measurement of percent tethering by high-throughput intact protein liquid chromatography–mass spectrometry (LC–MS) (Fig. 1b and Supplementary Table 1). The top 40 hits that demonstrated more than 2 s.d. above the mean percent tethering were further vetted by BME titration to calculate BME₅₀ values, which reflect the concentration of BME required to reduce percent tethering to 50% (Fig. 1b,c and Supplementary Table 2). Based on these analyses, the top tethering hits could be grouped into three structural categories: aromatic aza-heterocycles, fluoroalkyl benzamides and phenylacetamides, and diphenyl ethers (Supplementary Table 3). Overall, the aromatic aza-heterocycles exhibited the highest average percent tethering and contained biaryl pyrazole derivatives, in addition to indole and indazole analogs. Structure–activity analysis revealed that the benefit of additional bulk conferred by the bromine in the bromophenylpyrazole (1C18 [1], BME₅₀ = 905 μM) versus phenylpyrazole (1E4 [2], BME₅₀ = 466 μM) comparison could be partially compensated for by fusing the phenyl and pyrazole moieties into a smaller but more rigid indazole group (1A6 [3], BME₅₀ = 716 μM). The compound with the lowest BME₅₀ of this series (1C4 [4], BME₅₀ = 435 μM) consisted of an indole with a glyoxyl rather than a formyl amide linker, highlighting the importance of both spacing and hydrogen bonding in achieving effective small-molecule disulfide tethering to BAX. Ultimately, the top hit in the primary screen, 1C18 or N-(3-((2-aminoethyl) disulfanyl)propyl)-3-(4-bromophenyl)-1H-pyrazole-5-carboxamide (hereafter called Covalent BAX Inhibitor 1 or CBI1) (Supplementary Note 1), which also exhibited among the highest BME₅₀ values (905 μM), was selected for further study.

Fig. 1 ∣. Disulfide tethering identifies a covalent inhibitor of BAX.

a, Sensitization of BAX-mediated mitochondrial apoptosis by covalent reaction of t-2-hex (red) with BAX (gray, PDB ID: 1F16) at C126 (orange) inspired a disulfide tethering screen (bottom right) to identify small-molecule covalent modulators of BAX. b, A tethering screen of 1,600 disulfide-containing compounds against full-length, wild-type BAX yielded 40 fragments with percent tethering of at least 2 s.d. above the mean, as measured by MS. The top hit, CBI1, exhibited percent tethering that was 5 s.d. above the mean. c, Percent tethering upon BME titration at a fixed BAX:fragment ratio was measured for each of the top 40 hits from the tethering screen to validate binding and calculate BME₅₀ values. Thirty-two of the 40 compounds generated binding curves. d, Two-phase liposomal release assay evaluating the capacity of CBI1 to either activate or inhibit BAX in the presence of BME by measuring BAX-mediated liposomal release of fluorophore/quencher over time. Whereas tBID induced BAX-mediated liposomal poration (dark gray), the addition of CBI1 (50 μM) attenuated release (aqua). Data are mean ± s.e.m. for experiments performed in technical triplicate. e, Dose-responsive inhibition of tBID-triggered, BAX-mediated liposomal poration (dark gray) upon addition of CBI1 (blue). Data are mean ± s.e.m. for experiments performed in technical quadruplicate and repeated using independent preparations of liposomes and protein with similar results. WT, wild-type.

To determine if covalent derivatization of BAX by CBI1 exerts a functional effect, we performed a liposomal release assay designed to detect both activators and inhibitors of BAX-mediated permeabilization of fluorophore-encapsulated liposomes¹⁵, which simulate the membrane composition of the mitochondrial outer membrane. First, fluorescence monitoring was performed in the presence of BME for liposomes alone and liposomes plus BAX (wild-type, full-length) or the CBI1/BAX combination. Little to no liposomal release of fluorophore was detected (Fig. 1d), indicating that CBI1 was not an activator of BAX. After 50 min, the BH3-only protein tBID, a physiologic direct activator of BAX with no independent effect on liposomal poration (Supplementary Fig. 1 and Supplementary Data 1), was added to the BAX and CBI1/BAX conditions, and fluorescence monitoring was continued. Whereas tBID potently triggered BAX-mediated liposomal poration, the presence of CBI1 decreased both the kinetics and extent of fluorophore release (Fig. 1d). This inhibitory signature was then verified by dose-responsive blockade of BAX-mediated liposomal release by CBI1, with 50 μM CBI1 achieving complete inhibition (Fig. 1e). Notably, the aromatic aza-heterocycles 1A18 (5) and 1E4, which exhibited progressively lower percent tethering and BME₅₀ values, correspondingly demonstrated decreased capacity to inhibit BAX-mediated liposomal poration (Extended Data Fig. 1a-c and Supplementary Note 1). We further tested compound 3A20 (6) (similar BME₅₀ to CBI1) from the fluoroalkyl benzamide/phenylacetamide class of hits and observed minimal BAX-inhibitory activity in the liposomal release assay, suggesting that, in this case, binding does not yield appreciable functional inhibition (Extended Data Fig. 1d and Supplementary Note 1). Thus, the disulfide tethering screen identified a class of aromatic aza-heterocyclic compounds that both covalently derivatize BAX and inhibit its function with differential efficacy, with CBI1 demonstrating both the highest percent tethering and the most potent dose-responsive blockade of BAX-mediated liposomal poration upon triggering by the physiologic tBID ligand.

To verify that the inhibitory activity of CBI1 in the liposomal assay was specifically related to covalent reaction with BAX rather than tBID, which likewise has cysteines, we incubated tBID with excess CBI1 and observed no derivatization, as monitored by intact protein mass spectrometry (Supplementary Fig. 2 and Supplementary Data 1). 1n addition, we repeated the liposomal release experiments using a structurally stabilized and BAX-triggering BIM BH3 peptide helix that lacks cysteines and again found that CBI1 dose-responsively inhibited BAX-mediated liposomal poration (Extended Data Fig. 2). Whereas the above experiments demonstrated the capacity of CBI1 to inhibit BAX upon its initial activation by physiologic or synthetic BH3 engagement, we next examined whether CBI1 could block BAX self-association, a downstream step of the BAX activation pathway. For this experiment, we employed a ‘bypass’ reagent, specifically the detergent n-dodecyl-phosphocholine (Fos-12), which we previously reported transforms monomeric BAX into a stable and homogeneous BAX oligomer⁵. Pre-incubation of monomeric BAX with CBI1 did not disrupt Fos-12-induced oligomerization of BAX as monitored by size exclusion chromatography (SEC) (Extended Data Fig. 3), indicating that CBI1 acts at an upstream control point, specifically the conformational activation step of ligand-triggered monomeric BAX.

CBI1 targeting of C126 influences a key region of BAX

BAX contains two cysteine residues: C62, which is a residue of the BH3 helix (α2) and only partially exposed on the protein surface, and C126, which is located at the junction of the α5/α6 hairpin and fully surface exposed (Fig. 2a). To determine whether CBI1 reacts with C62, C126 or both cysteine residues, we generated C62A and C126A mutants of BAX and then incubated CBI1 with either wild-type BAX or each of the cysteine mutants upon titration of BME and then determined percent tethering by intact protein mass spectrometry. Whereas C126A mutagenesis abrogated CBI1 derivatization of BAX across all BME concentrations, C62A mutagenesis had no such effect and demonstrated the identical percent tethering curve as wild-type BAX (Fig. 2b). To determine whether CBI1 could achieve this level of BAX cysteine selectivity in the context of a complex protein mixture, we incubated CBI1 with K562 cell lysate spiked with recombinant wild-type BAX (1 μM) and monitored for a dose-responsive decrease in cysteine-containing peptides–as a result of CBI1 derivatization–by mass spectrometry. Whereas detection of the BAX peptide containing C126 was dose-responsively decreased upon CBI1 treatment, detection of the BAX peptide containing C62 was unaffected (Supplementary Fig. 3 and Supplementary Data 1). Although the disulfide warhead of CBI1 precludes its use in cells due to the reducing environment, we noted that the compound reacted with only 0.4% of the possible cysteine-containing peptides of the lysate upon filtering for a threefold reduction of peptide detection in the presence of CBI1 at a significance threshold of P = 0.01, with the BAX C126 peptide among this 0.4%.

Fig. 2 ∣. CBI1 targets BAX at C126.

a, BAX (PDB ID: 1F16) contains two potential sites for CBI1 derivatization: C62 and C126. b, Comparative percent tethering of CBI1 to BAX wild-type (gray), C62A (purple) or C126A (orange) upon titration of BME revealed that C126A mutagenesis abrogated CBI1 covalent derivatization of BAX, whereas C62A mutagenesis had no effect. Data are mean ± s.d. for experiments performed three times using independent preparations of proteins. c, Representative lowest-energy pose from the molecular dynamics simulation of CBI1-derivatized BAX. CBI1 (cyan), BAX (gray; PDB ID: 1F16). d, Contact probabilities for BAX–CBI1 interactions calculated as the fraction of the 200-ns simulation that (at least) one atom of a BAX residue is within 5 Å of CBI1. e, Residues with a contact probability of 0.6 or greater (d) are mapped onto the ribbon diagrams of BAX (PDB ID: 1F16). WT, wild-type.

To predict how CBI1 binds to BAX, we performed molecular dynamics simulations of BAX upon disulfide tethering of CBI1. Using this computational approach, we can examine in an unbiased fashion how CBI1 may engage with wild-type BAX upon covalent derivatization. A lowest-energy, stable conformation demonstrated the disulfide bond between BAX C126 and CBI1, in addition to a series of residues with high contact probabilities in distal α1 (D33, R34 and A35), distal α2 (D68 and D71), α5 (K119, V121, L122, K123, A124 and L125) and the α5/α6 junction (T127 and K128) (Fig. 2c,d and Supplementary Video). The proposed interacting residues form a binding site on the BAX surface, with the molecule appearing to reinforce a nexus of interactions among α-helices 1, 2 and 5 (Fig. 2e). Because mobilization of the α1–α2 loop, exposure of the hydrophobic surface of BAX BH3 (α2) and destabilization of core interactions between α5 and its surrounding helices are essential components of the BAX activation mechanism, stabilization of this region by CBI1 could explain the small molecule’s inhibitory activity.

To vet this mechanistic hypothesis, we performed heteronuclear multiple quantum coherence (HMQC) NMR analysis of ¹⁵N-BAX upon addition of CBI1. Such NMR-based structural analyses of BAX have previously been applied to localize both activating and inhibitory ligand-binding sites and the conformational consequences of interaction^3,15-18. Upon addition of CBI1 at small molecule:protein ratios of 5:1 and 10:1, we observed a series of chemical shift changes that localize to the immediate vicinity of C126, consistent with selective molecular engagement of C126, in addition to key surrounding residues (Fig. 3a, Supplementary Fig. 4a,b and Extended Data Fig. 4a). At the 5:1 condition, we observed two distinct cross-peaks for each of a series of affected residues, indicative of slow exchange between the CBI1-bound and unbound conformations of BAX, with conversion to one cross-peak each upon complete BAX C126 derivatization at the 10:1 condition (Extended Data Fig. 4b,c). Mapping the chemical shift changes of the 10:1 condition onto monomeric BAX, including those residues whose cross-peaks demonstrated prominent signal attenuation or chemical shift perturbation, highlighted the impacts of CBI1 derivatization both in the immediate vicinity of C126, consistent with the binding site predicted by the molecular dynamics simulation (Fig. 2c-e), and in key adjacent regions that undergo allosteric sensing. Specifically, in addition to identifying changes in the core α5 helix that contains C126, we observed perturbations at the confluence of α5 interactions with the α3–α4 loop, α6 and α9 (Fig. 3b). On the opposite side of α5 to this α3/α4/α6/α9 nexus, discrete residues of α1, the α1–α2 loop and α2 were also affected, a region critical to the initiation and propagation of BAX activation. Indeed, a difference distance matrix plot derived from molecular dynamics simulations of wild-type BAX in the presence or absence of CBI1 demonstrated the greatest effect of small-molecule derivatization on the dynamics of the α1–α2 loop (Extended Data Fig. 5a), consistent with prominent chemical shift perturbation of loop residues G39, G40 and L45, as detected by HMQC NMR (Fig. 3). Comparison of the α1–α2 loop position in the molecular dynamics simulations of BAX in the presence or absence of CBI1 derivatization revealed newfound interaction of the loop in the vicinity of small-molecule binding (Extended Data Fig. 5b,c), which may serve to restrict loop mobilization and, thus, exposure of the critical α2 (BH3) helix. Indeed, we previously observed that covalent tethering of the α1–α2 loop to α6 via an installed disulfide bridge between amino acid positions 45 and 137 blocked the conformational activation of BAX⁴. F116 of the BAX α5 core, a residue that we recently reported to be essential to stabilizing the conformationally inactive state of monomeric BAX⁶, also exhibited among the most prominent chemical shift changes (Fig. 2). Taken together, our mass spectrometry, molecular dynamics simulation and NMR data support a mechanism by which selective derivatization of C126 by CBI1 stabilizes a series of interactions between α5 and key structures that regulate the conformation of monomeric BAX.

Fig. 3 ∣. NMR analysis of the BAX–CBI1 interaction.

a, Measured chemical shift changes of ¹⁵N-BAX (20 μM) upon addition of CBI1 (10:1 of CBI1:BAX, 2% DMSO), plotted as a function of BAX residue number. Chemical shift changes above the 2 s.d. cutoff (significance threshold of 0.0572 p.p.m.) are colored maroon (G40, R65, A82, T85, F116, S118 and T127), and those above the 1 s.d. cutoff (significance threshold of 0.0373 p.p.m.) are colored red (G39, V83, N106, V111, R134 and C126). Residues whose cross-peaks demonstrated prominent signal attenuation or chemical shift perturbation upon CBI1 incubation are colored beige (Q32, A35, L45, L70, A81, E90, L125, K128, T135, T140, L141, V180, L181, M191 and G192). b, Residues represented in a as maroon, red and beige bars are mapped onto the ribbon diagrams of BAX (PDB ID: 1F16).

Next, to evaluate the impact of CBI1’s covalent derivatization of BAX C126 relative to its non-covalent interactions in the region surrounding C126, we repeated the NMR analyses, this time comparing the chemical shift perturbations induced in (1) wild-type BAX by CBI1 and a non-covalent analog of CBI1, termed N-CBI1 (7), in which the S-S moiety of CBI1 is replaced by C-C (Supplementary Note 1), and in (2) wild-type BAX versus BAX C126A upon exposure to CBI1. In both cases, whether by eliminating the covalent reactivity of the molecule or by C126A mutagenesis of BAX, we observed a marked decrease in the magnitude of chemical shift changes; whereas key features of the NMR signature of wild-type ¹⁵N-BAX upon CBI1 titration remain, it is notable that changes in the α1–α2 loop are no longer apparent, suggesting that restraint of the loop may be diminished in the absence of covalent derivatization (Supplementary Fig. 4b-d and Extended Data Fig. 6a-c). Also of interest, the slow exchange between bound and unbound states observed for covalent engagement, as reflected by the presence of two discrete cross-peaks (or marked signal attenuation or chemical shift perturbation) for affected residues, is replaced by stepwise migration of cross-peaks indicative of the fast exchange that occurs upon non-covalent interaction (Extended Data Fig. 4b-e). These data underscore the key influence of covalent derivatization in driving the direct and allosteric chemical shift perturbations observed in BAX. In addition, non-covalent binding, albeit weaker, is independently evident, which is consistent with the very mechanism of disulfide tethering that selects for small molecules with non-covalent interactions immediately adjacent to the site of thiol-disulfide exchange.

CBI1 restrains the auto-activity of BAX F116A

We recently reported that BAX residues 113–116, which are radially distributed along a single turn of the core α5 helix, collectively engage with every other α-helix of the BAX structure, with alanine mutagenesis of any one residue resulting in varying degrees of BAX auto-activation or sensitization of ligand-triggered BAX activation⁶. Of these mutants, BAX F116A emerged as the most auto-active in vitro and in cells. Given the prominent chemical shift perturbation that occurs in F116 upon CBI1 derivatization of BAX C126 (Fig. 3), we tested the capacity of CBI1 to functionally restrain BAX F116A, as measured by liposomal release assay. We observed that CBI1 effectively suppressed the auto-activity of BAX F116A, as reflected both by the maximal extent and by the kinetics of liposomal release, with 1.5 μM dosing causing near-complete inhibition (Fig. 4a-c). We next sought to explore the relative contributions of CBI1’s covalent and non-covalent interactions to the inhibition of BAX F116A auto-activation in the presence of liposomes. In examining the comparative influence of CBI1 and N-CBI1 on both BAX F116A and BAX F116A/C126A, we observed that CBI1 exhibited superior inhibitory potency than N-CBI1 in the presence of C126 but was equipotent to N-CBI1 upon C126 mutagenesis (Extended Data Fig. 7a,b). These results again highlight a key mechanistic attribute of the disulfide tethering strategy in that an inherent feature of top covalent hits is concomitant non-covalent engagement, which facilitates selective cysteine targeting. Analogous to what we observed by NMR, eliminating the capacity to form a covalent bond, whether by replacing the S-S moiety of CBI1 with C-C in N-CBI1 or mutating BAX C126 to an alanine, correspondingly abrogates the potency advantage conferred by covalent reactivity.

Fig. 4 ∣. CBI1 suppresses the membrane-permeabilizing auto-activity of BAX F116A.

a–i, CBI1 dose-responsively inhibited the auto-activity of BAX F116A, as monitored by BAX-mediated liposomal release of fluorophore/quencher over time (a) and quantified by maximum fraction release (b) and area under the curve (c). In contrast, BAX inhibitors BAI1 (d–f) and EO (g–i) had little to no effect on the auto-activity of BAX F116A. Data are mean ± s.e.m. for experiments performed in technical quadruplicate and repeated twice more using independent preparations of liposomes and protein with similar results. j, Mitochondrial cytochrome $c$ release from BAX/BAK-deficient mouse liver mitochondria upon treatment with BAX F116A was measured in the presence or absence of CBI1, BAI1 or EO. BAX F116A-mediated cytochrome $c$ release (dark gray) was dose-responsively suppressed by CBI1 (blue) but not by BAI1 (green) or EO (purple). Data are mean ± s.e.m. for experiments performed in technical quadruplicate and repeated twice more using independent preparations of mitochondria and protein with similar results.

We next tested whether non-covalent inhibitors of BAX that operate at distinct sites from CBI1 could restrain BAX F116A in the liposomal release assay. Whereas BAI1 targets a unique site centered at the convergence of α3, α4, α5 and α6 (ref. 17), eltrombopag (EO) engages the BH3 trigger site¹⁹. In each case, BAI1 or EO had little to no inhibitory effect on BAX F116A auto-activity (Fig. 4d-i). To extend the analysis to the physiologic context of the mitochondrial membrane, we performed comparative cytochrome $c$ release assays using mitochondria purified from Bax/Bak-null mouse livers. Like the liposomal release assay results, CBI1 (0.2–2 μM dose range) suppressed BAX F116A-mediated cytochrome $c$ release, but BAI1 and EO had little to no inhibitory effect (Fig. 4j).

To investigate the mechanism by which CBI1 inhibits BAX F116A auto-activity, we performed hydrogen deuterium exchange mass spectrometry (HDX MS), examining the influence of small-molecule derivatization of BAX C126 on the conformation of BAX F116A. HDX MS probes protein structure by measuring the deuterium incorporation of backbone amides²⁰. When diluted into deuterium buffer, backbone hydrogens of flexible and/or exposed protein regions rapidly exchange with deuterium, whereas buried domains and/or those regions that contain hydrogen bonding involving backbone amide hydrogens demonstrate slowed or suppressed deuterium exchange. We first tested the effect of CBI1 on wild-type BAX and observed no significant differences in the deuterium exchange profile in the presence or absence of the small molecule, consistent with wild-type BAX already existing in its conformationally stable, inactive state (Supplementary Fig. 5 and Supplementary Data 1). In comparing the deuterium exchange profiles of wild-type and F116A BAX⁶, we observed that alanine mutagenesis resulted in early deprotection of α9 and progressive deprotection over time of the distal region of α2, in addition to α3, α4, and α6 (Fig. 4a). This conformational exposure affects key functional domains of BAX, including the junction of α3 with the critical BH3 α2 helix, the membrane-lytic α6 helix and the C-terminal mitochondrial translocation α9 helix. Strikingly, CBI1 derivatization of BAX F116A at C126 completely reversed these conformational changes (Fig. 5b), as evidenced by the deuterium difference profile of BAX F116A/CBI1 versus BAX F116A appearing as the near mirror image of that of BAX F116A versus BAX wild-type (Fig. 5a,b). Likewise, a difference distance matrix plot derived from molecular dynamics simulations of BAX F116A in the presence or absence of CBI1 demonstrated striking reversal of the auto-activating conformational changes of BAX F116A upon CBI1 covalent derivatization of C126 (Extended Data Fig. 8). Consistent with the liposomal and mitochondrial cytochrome $c$ release data, incubation of BAX F116A with BAI1 or EO had no effect on the deuterium exchange profile of BAX F116A (Supplementary Fig. 6 and Supplementary Data 1). Interestingly, in comparing the deuterium exchange profile of BAX F116A/CBI1 with that of BAX wild-type, it is noteworthy that CBI1 treatment essentially restores wild-type conformational dynamics to the BAX F116A mutant (Fig. 5c). The only distinction between the two deuterium exchange profiles lies in those portions of α3 and α9 that directly face CBI1-derivatized C126, with CBI1 inducing the protection of these regions even beyond that observed for the inactive, wild-type BAX monomer (Fig. 5c). Thus, our biochemical, HDX MS and molecular dynamics simulation data indicate that CBI1 is uniquely effective at blocking the auto-activity of BAX F116A by imposing marked conformational constraint, which restores BAX F116A to the inactive state of monomeric, wild-type BAX.

Fig. 5 ∣. CBI1 reverses the conformational activation of BAX F116A.

a–c, Deuterium difference plots showing the relative deuterium incorporation of BAX F116A minus that of wild-type BAX (a), BAX F116A in the presence of CBI1 (25:1 of CBI1:BAX) minus that of BAX F116A alone (b) and BAX F116A in the presence of CBI1 (25:1 of CBI1:BAX) minus that of wild-BAX (c), as measured at 10 s, 1 min and 10 min of deuterium labeling. Regions of deprotection (orange) and protection (green) above the 0.5-Da significance threshold (shown as gray shading on each plot) at 10 min of labeling are mapped onto the solution structure of BAX (PDB ID: 1F16), with C126 and F116 highlighted in red and magenta, respectively. HDX MS experiments were performed at least twice using independent preparations of BAX proteins. WT, wild-type; ND, no data.

CBI1 blocks t-2-hex lipidation and oligomerization of BAX

Given the role of t-2-hex in sensitizing the activation of BAX, we sought to determine if, in addition to imposing conformational constraint, CBI1 could further suppress BAX activation by competitive inhibition of C126 lipidation. Whereas the initiation of BAX activation occurs in the cytosol, lipidation and homo-oligomerization occur at the mitochondrial outer membrane. To distinguish between the potential dual components of a mechanism that inhibits BAX-mediated mitochondrial cytochrome $c$ release, we first examined the influence of CBI1 derivatization of BAX C126 on mitochondrial translocation. Specifically, we incubated BAX F116A with CBI1 in solution, followed by the addition of mouse liver Bax/Bak-null mitochondria and monitored mitochondrial translocation by BAX western analysis of the isolated supernatant and pellet fractions. Consistent with its suppression of BAX F116A-mediated liposomal and cytochrome $c$ release (Fig. 3), and the conformational constraint imposed on BAX F116A as delineated by HDX MS (Fig. 4), CBI1 dose-responsively suppressed the auto-translocation of BAX F116A from the solution phase to mitochondria (Extended Data Fig. 9a). Indeed, the mitochondrial translocation of BAX requires dissociation of the α9 helix from the binding groove at the C-terminal face, a conformational dynamic specifically reversed by CBI1 (Fig. 5b,c). Of note, the inhibitory effect of CBI1 on BAX F116A auto-translocation was essentially eliminated upon C126A mutagenesis, again highlighting the mechanistic importance of CBI1 derivatization of C126 (Extended Data Fig. 9b).

Small-molecule inhibition of BAX lipidation can potentially occur indirectly, by blocking mitochondrial translocation, and/or directly, by competitive covalent reaction at C126. Because CBI1 can indeed suppress the mitochondrial translocation of BAX F116A (Extended Data Fig. 9a), which could confound a determination of whether any decrement in lipidation occurs by indirect or direct means, we performed an assessment of the lipidation reaction in isolation. Specifically, we exposed BAX F116A to t-2-hex in the presence or absence of increasing amounts of CBI1 and then monitored BAX lipidation and resultant homo-oligomerization, as detected by the BAX fluorescence induced upon addition of Cy5-hydrazide, which reacts with the aldehyde of t-2-hex on lipidated BAX⁸. Gel electrophoresis and fluorescence scan of the mixtures revealed that CBI1 dose-responsively blocked the lipidation and homo-oligomerization (as reflected by laddering) of BAX F116A (Extended Data Fig. 10). Indeed, protein staining for the 21-kDa BAX monomer revealed little to no residual monomeric species in the t-2-hex-treated sample but dose-responsive restoration of detectable monomeric BAX upon addition of CBI1 (Extended Data Fig. 10). Analogous to the liposomal release assay findings (Extended Data Fig. 7a,b), the covalent feature of CBI1 enabled more potent blockade of t-2-hex activity as compared to the non-covalent analog, N-CBI1 (Extended Data Fig. 10).

To further confirm that the inhibitory activity of CBI1 and the mechanistic insights gleaned from our analyses of its influence on BAX F116A mitochondrial translocation, t-2-hex lipidation and mitochondrial cytochrome $c$ release are relevant to the physiologic BAX activation pathway, we repeated the battery of analyses using wild-type BAX in conjunction with its endogenous trigger, tBID. First, we demonstrated the capacity of CBI1 to dose-responsively impair the mitochondrial translocation of wild-type BAX upon tBID triggering, with little to no inhibitory effect observed upon C126A mutagenesis of wild-type BAX (Fig. 6a-d). Second, CBI1 inhibited the lipidation and homo-oligomerization of wild-type BAX in both a dose-dependent and a time-dependent fashion (Fig. 6e). Third, as observed in the context of BAX F116A, eliminating the covalent feature of CBI1 weakened the capacity of N-CBI1 to block lipidation and homo-oligomerization (Fig. 6f). Finally, CBI1 dose-responsively inhibited tBID-triggered mitochondrial cytochrome $c$ release by wild-type BAX, an effect that could be further enhanced when CBI1 was combined with BAI1, a non-covalent BAX inhibitor that operates at a distinct site (Fig. 6g,h). Taken together, these data demonstrate that CBI1 is capable of blocking BAX not only by imposing conformational constraint but also by suppressing t-2-hex-mediated sensitization of homo-oligomerization through competitive blockade of lipidation at C126.

Fig. 6 ∣. CBI1 inhibits BAX by a dual mechanism of action.

a–d, Distribution of BAX wild-type (WT) (a) or BAX C126A (b) (2 μM) between supernatant and BAX/BAK-deficient mitochondrial fractions, as detected by BAX western analysis after pre-incubating BAX proteins with CBI1 (25 μM–100 μM, lanes 3–5), addition to mitochondria in the presence of tBID (40 nM), isolation of the supernatant and pellet fractions by centrifugation, and SDS-PAGE. Isolation of the mitochondrial pellet fraction was verified by VDAC1 western analysis. Fraction of BAX in the pellet was quantified by densitometry (c,d). The experiment was performed twice using independent preparations of protein, mitochondria and molecule. e, Pre-incubation of BAX WT (5 μM) with 12.5 μM or 25 μM CBI1 for 5 min, 30 min or 60 min, followed by treatment with t-2-hex (2.5 mM) for 2 h at 37 °C and detection of lipidated BAX WT by addition of Cy5-hydrazide, gel electrophoresis and fluorescence scan. Derivatization and induced homo-oligomerization (laddering) of BAX WT by t-2-hex, and the dose-responsive and time-responsive suppression by CBI1, was detected by fluorescence scan of the BAX mixtures (top and middle panels). Effect of treatment on monomeric BAX levels was monitored by protein stain (bottom panel). The experiment was performed twice using independent preparations of protein, lipid and molecule. f, Incubation of BAX WT (5 μM) and t-2-hex (2.5 mM) in the presence or absence of escalating doses (5 μM–100 μM) of CBI1 (left) or N-CBI1 (right), followed by detection of lipid derivatization, induced homo-oligomerization and monomeric BAX protein as described in e. The experiment was performed twice using independent preparations of protein, lipid and molecules. g,h, Cytochrome $c$ release from BAX/BAK-deficient mitochondria upon treatment with BAX WT, tBID, CBI1 or BAX WT incubated with increasing doses of CBI1 before adding tBID (g) or with BAX WT, tBID, CBI1, BAI1 or BAX WT treated with combinations of CBI1 and/or BAI1 before adding tBID (h), as measured by ELISA. Data are mean ± s.e.m. for experiments performed in technical triplicate (g) or quadruplicate (h) and repeated using independent preparations of mitochondria, protein and molecules with similar results. WB, western blot.

Discussion

Given the pivotal role of BAX as an executioner protein of the cell death pathway, pharmacologic modulation of its pro-apoptotic activity has the potential to treat a host of human diseases by restoring the balance between cellular life and death²¹. Mechanistic insights into the physiologic regulation of BAX have provided blueprints for the development of peptide and small-molecule prototypes that target BAX. For example, the discovery of the BAX α1/α6 trigger site that mediates direct activation of BAX by select BH3-only proteins and their corresponding hydrocarbon-stapled peptides^3,22 inspired the pursuit of an in silico screen, which identified the first direct and selective BAX activator molecules (BAMs)²³. Iterative optimization of BAM7 yielded BAX trigger site activators 1 and 1.2 (BTSA1 and BTSA1.2), which demonstrated anti-leukemic activity in a mouse xenograft model of human acute myeloid leukemia²⁴ and synergy with navitoclax in patient-derived xenograft models of human colorectal cancer²⁵, respectively. Whereas physiologic inhibitors of monomeric BAX have been described, such as the BH4 domain of BCL-2 (ref. 16) and the vMIA protein of cytomegalovirus¹⁸, the discovery of small molecules that block BAX has largely relied on functional screens^17,26.

The identification of t-2-hex as a sensitizer of BAX activation at the mitochondria⁷ inspired us to investigate the mechanism of action, which could potentially be exploited for therapeutic modulation of BAX. We demonstrated that t-2-hex, a lipid electrophile, selectively derivatized BAX C126 by non-enzymatic Michael addition, potentiating the activation of BAX⁸. Here, by use of a disulfide tethering screen, we identified and characterized what is, to our knowledge, the first small-molecule covalent modulator of BAX. CBI1 selectively reacts with BAX at C126, but, rather than phenocopy the sensitization activity of t-2-hex, it inhibits BAX activation induced by BH3 ligands, F116A mutagenesis or t-2-hex. Indeed, the contrasting effects of t-2-hex and CBI1 on BAX activation underscores that covalent targeting of C126 in and of itself does not determine a gain-of-function or loss-of-function outcome. Instead, how the derivatizing ligand engages the structural components surrounding the target cysteine and the conformational consequences are determinative, as also seen in a disulfide tethering screen of PDK1 where disulfide fragment hits included both activators and inhibitors²⁷. HDX MS revealed that t-2-hex lipidation induced early deprotection of α1, α1–α2 loop and α9 (ref. 8), key regions implicated in the initiation of BAX activation and mitochondrial translocation, respectively⁴. In contrast, CBI1 restrained an auto-active mutant of BAX⁶ by stabilizing core interactions between α5 and its surrounding residues, with molecular dynamics simulations indicating that covalent engagement of BAX C126 by CBI1 restricts rather than mobilizes the α1–α2 loop of wild-type BAX. Furthermore, we previously reported that lipid electrophiles with shorter chain lengths than t-2-hex were more reactive toward BAX yet less capable of inducing BAX oligomerization⁸. These findings suggest that the precise non-covalent binding mode of C126-reactive ligands dictates their functional effect on BAX activity, a phenomenon also observed at other regulatory sites on BAX. For example, transient binding of stapled BH3 peptides and BAMs/BTSAs at the α1/α6 trigger site initiates BAX activation^3,23,24, whereas stable engagement of the site by EO via a distinct non-covalent binding orientation results in inhibition of BAX activation¹⁹. A discrete surface pocket formed by the flexible loops between helices α1–α2, α3–α4 and α5–α6 and a portion of the C-terminal α9 helix is the binding site for both the inhibitory cytomegalovirus vMIA peptide¹⁸ and BIF-44, a small molecule that sensitizes BAX activation¹⁵. In the present study, we found that CBI1 not only competes for C126 derivatization to block t-2-hex-mediated BAX sensitization but also independently restrains BAX by stabilizing key structures of its monomeric architecture.

BAX has long been established as a critical mediator of pathologic cell death in diseases of stress-induced cell loss, such as heart attack, stroke and the spectrum of neurodegenerative diseases^28-31. Relyvrio was recently approved by the FDA for the treatment of amyotrophic lateral sclerosis (ALS) and is thought to achieve neuroprotection, at least in part, by suppression of BAX^32,33. Relyvrio is a two-component drug composed of sodium phenylbutyrate and ursodoxicoltaurine (TUDCA). TUDCA, a hydrophilic bile acid, is thought to alter the mitochondrial outer membrane composition in a manner that impedes BAX insertion, thereby preserving mitochondrial function and suppressing apoptosis³⁴. Whereas TUDCA inhibits BAX indirectly, molecules that directly bind and inhibit BAX activation could provide a more targeted therapeutic approach for ALS and perhaps other neurodegenerative diseases. BAI1 and EO are two such molecules that were reported to suppress the activation of BAX through direct interaction, resulting in the blockade of cell death^17,19,35. We found that covalent targeting of BAX C126 provides an alternative and distinct mechanism of direct BAX inhibition. Not only does CBI1 effectively compete with the physiologic t-2-hex ligand that otherwise sensitizes BAX activation, but it also stabilizes the inactive conformation of BAX in a regiospecific manner not achieved by BAI1 or EO. Thus, our results provide a compelling rationale for the advancement of small-molecule covalent inhibitors that target BAX C126 to suppress BAX-mediated apoptosis in diseases of premature or unwanted cell death.

Methods

Compounds

tBID (883-M8-050, R&D Systems), BAI1 (S8865, Selleck Chemicals), EO (S4502, Selleck Chemicals), t-2-hex (17566, Cayman Chemical) and Fos-12 (F308, Anatrace) were purchased from the indicated vendors.

Recombinant BAX expression and purification

Recombinant full-length, wild-type BAX in the pTYB1 vector was expressed in BL21 (DE3) Escherichia coli as previously reported^2,3,15. The F116A and F116A/C126A mutants of BAX were generated by PCR-based mutagenesis (Q5 Site-Directed Mutagenesis Kit, New England Biolabs (NEB)) (Supplementary Table 4) and confirmed by DNA sequencing. Transformed E. coli were cultured in LB medium containing carbenicillin (0.1 g L⁻¹) and grown to an optical density (OD) of 0.6–0.8, and protein expression was induced by the addition of 1.0 mM IPTG at 30 °C for 4 h. Bacterial pellets were resuspended in lysis buffer (20 mM Tris, 250 mM NaCl, pH 7.2) containing protease inhibitor tablets (Roche) and lysed over three passages through a microfluidizer (Microfluidics) on ice. The soluble fraction was isolated by centrifugation at 48,384g for 45 min at 4 °C. BAX protein was purified by chitin affinity chromatography using chitin resin (NEB) on a gravity flow column. The intein and affinity tag were cleaved using 10 mg ml⁻¹ dithiothreitol at 4 °C for 16–60 h. The full-length, tagless protein was eluted, concentrated and purified by SEC in fast protein liquid chromatography (FPLC) buffer (20 mM HEPES-KOH, 150 mM KCl, pH 7.2) using a Superdex S-75 (GE Healthcare) column on an FPLC system (ÄKTA pure, GE Healthcare Life Sciences). Protein purity and identity were confirmed by Coomassie stain and western analysis using the 2D2 mouse monoclonal BAX antibody (Santa Cruz Biotechnology, sc-20067; RRID: AB_626726; 1:200). BAX C62A and C126A mutants were generated by PCR-based mutagenesis (Q5 Site-Directed Mutagenesis Kit, NEB) and confirmed by DNA sequencing. Overnight cultures were grown in MDG-carbenicillin, and bacterial colonies were plated onto MDG-carbenicillin agar plates³⁶. Highly expressing clonal populations were selected by double colony selection³⁷. The resulting colonies were grown overnight in MDG-carbenicillin, inoculated into LB, grown to OD 0.6–0.8 and induced with 1.0 mM IPTG overnight at 16 °C. Recombinant protein was then isolated and validated as described above for wild-type, F116A and F116A/C126A BAX.

Disulfide tethering fragment screening by mass spectrometry

A 1,600-member disulfide fragment library^13,14 was screened by incubating recombinant, full-length, wild-type BAX (1 μM) with each compound individually (100 μM) for 1 h at room temperature in BAX FPLC buffer containing BME (500 μM) in 384-well plates. Each well was then screened by high-throughput intact protein LC–MS for a change in the mass of BAX that corresponded to the individual fragment. Mass spectra were deconvoluted using MagTran1.03b2 software. The fragments were then ranked based on percent tethering to BAX, calculated as $({\mathrm{B}}_{\mathrm{F}}∕{\mathrm{B}}_{\text{tot}})\times 100$, where ${\mathrm{B}}_{\mathrm{F}}$ is the abundance of fragment-modified BAX, and ${\mathrm{B}}_{\text{tot}}$ is the sum of the abundance of unmodified, BME-modified and fragment-modified BAX. The top 40 fragments, which were more than 2 s.d. above the mean percent tethering for the 1,600 fragments tested, were advanced to binding validation by BME titration. BME titration experiments were performed by incubating the fragments with wild-type BAX as above while varying BME concentration, followed by intact protein LC–MS analysis. Percent tethering of the fragments to wild-type BAX at varying BME concentrations was calculated as described above, and nonlinear regression analysis of dose–response curves was performed using Prism 9 software (GraphPad). The concentration of BME required to reduce tethering to 50% (BME₅₀) was calculated by interpolation in Prism 9 software.

Liposomal release assay

Large unilamellar vesicles (LUVs) encapsulating the fluorophore/quencher pair 8-aminonapthalene-1,3,6-trisulfonic acid (ANTS) and p-xylene-bis-pyridinium bromide (DPX) were formed by liposome extrusion and purified by SEC. In brief, a lipid mixture that mimics the composition of the mitochondrial outer membrane was generated by dissolving a 48:28:10:10:4 molar ratio of phosphatidylcholine, phosphatidylethanolamine, phosphatidylinositol, dioleoyl phosphatidylserine and tetraoleolyl cardiolipin (Avanti Polar Lipids) in chloroform. To produce lipid films, nitrogen gas was used to evaporate the chloroform, and the lipids were dried under high vacuum overnight. For long-term storage, lipid films were maintained in a nitrogen atmosphere at −80 °C. Lipid films were resuspended in 1 ml of liposomal release assay buffer (20 mM HEPES-KOH, 150 mM KCl, 5 mM MgCl₂, pH 7.2) containing the fluorophore/quencher pair ANTS (12.5 mM) and DPX (45 mM). Liposomes were formed by exposing the resuspended lipid mixture to five freeze–thaw cycles and then extruding the liposomes using a 100-nm polycarbonate membrane 11 times. Liposomes were then purified from unencapsulated ANTS and DPX using a Sepharose CL-2B size exclusion column (GE Healthcare). The liposomal release assay was performed using an M1000 Infinite (Tecan) equipped with i-control 1.11 software (Tecan). Recombinant BAX proteins were pre-incubated with vehicle (0.0675–0.75% DMSO) or compounds (CBI1, N-CBI1, BAI1, EO, 1A18, 1E4 and 3A20) in liposomal release assay buffer for 1 h at room temperature and then added to liposomes (final concentration BAX = 500 nM; final concentrations of compounds as indicated in the figures), and ANTS fluorescence (355-nm excitation, 540-nm emission and 20-nm slit width) was monitored every 60 s after the addition of tBID (final concentration 25 nM) or BIM SAHB³ (final concentration 250 nM) to the appropriate wells. For assessment of the independent effect of tBID on liposomal poration, tBID was added to liposomes (final concentrations of tBID = 25 nM–200 nM) in the presence or absence of BAX wild-type (final concentration BAX = 500 nM), and ANTS fluorescence (355-nm excitation, 540-nm emission and 20-nm slit width) was monitored every 60 s. For the two-phase liposomal release assay, wild-type BAX (500 nM) was equilibrated with or without CBI1 (50 μM) and BME (250 μM) in the presence of liposomes for 50 min at room temperature, with ANTS fluorescence measured every 60 s. After the equilibration period, tBID (10 nM) was added to the appropriate wells, and ANTS fluorescence was measured every 60 s for an additional 86 min. Maximal release was determined by the addition of Triton X-100 to a final concentration of 0.625% (v/v). Fraction release was calculated as $((\mathrm{F}-{\mathrm{F}}_0)∕({\mathrm{F}}_{100}-{\mathrm{F}}_0))$, where $\mathrm{F}$ is the observed fluorescence at a given time, and ${\mathrm{F}}_0$ and ${\mathrm{F}}_{100}$ represent baseline and maximal fluorescence, respectively. The data were analyzed using Microsoft Excel 2023 software and plotted with Prism 9 software. Maximum fraction release and area under the curve values were determined in Prism 9 software.

Analysis of oligomeric BAX by SEC

Monomeric wild-type BAX (20 μM) was incubated in the presence or absence of CBI1 (200 μM) for 1 h at room temperature. The detergent Fos-12 was then added (final concentration 3 mM) to the appropriate samples, and the reactions were incubated for 16 h at 4 °C. The samples were then concentrated (10 kD spin concentrator, Millipore) and loaded onto a Superose 6 Increase size exclusion column (GE Healthcare) equilibrated with BAX FPLC buffer containing 3 mM Fos-12.

Intact protein mass spectrometry of tBID upon CBI1 incubation

tBID (2 μM) was incubated in the presence or absence of CBI1 (50 μM) in liposomal release assay buffer for 1 h at room temperature. The samples were then injected onto a self-packed reversed-phase column (500 μm ID, 4 cm of POROS 50R2 resin). After desalting for 4 min, protein was eluted using an HPLC gradient of 0–100% Solution B over 1 min at a flow rate of 30 μl min⁻¹ (Solution A, 0.2 M acetic acid in water; Solution B, 0.2 M acetic acid in acetonitrile) into an LTQ ion trap mass spectrometer (Thermo Fisher Scientific) that acquired profile MS spectra (m/z 300–2,000). Mass spectra were deconvolved using MagTran1.03b2 software.

Intact protein mass spectrometry of BAX cysteine mutants upon CBI1 incubation

CBI1 (80 μM) was incubated with wild-type, C62A or C126A BAX (800 nM) in BAX FPLC buffer (see above) with a dilution series of BME for 1 h at room temperature. The reactions were quenched by fivefold dilution with 0.1% formic acid in water. For each mass spectrometry analysis, protein was injected onto a self-packed, reversed-phase trap column (9.5 × 30 mm, POROS 20 R2 resin). After desalting for 3 min, protein was eluted using a 3-min 15–70% gradient of acetonitrile with 0.1% formic acid, at a flow rate of 100 ml min⁻¹. Eluant was directed into a Xevo G2 electrospray mass spectrometer where spectra were recorded over m/z range 500–2,000 with a 1.0-s scan rate. Mass spectra were deconvoluted using MagTran1.03b2 software. The percent tethering of CBI1 to each BAX construct at varying BME concentrations was calculated from the unmodified and modified masses, as described for the disulfide tethering fragment screen. Nonlinear regression analysis of dose–response curves was performed using Prism 9 software.

Chemoproteomic analysis of BAX cysteine derivatization In-lysate reactive cysteine profiling.

K562 cells were grown in RPMI 1640 (Corning) medium supplemented with 10% FBS and 1% penicillin–streptomycin (Gibco) to near confluence, collected, washed twice with cold PBS and stored at −80 °C until lysis. Frozen cells were resuspended with lysis buffer (PBS, pH 7.4, 0.1% NP-40) using a syringe equipped with a 21-gauge needle. The crude lysate was then subjected to an EpiShear Probe Sonicator (Active Motif) for further homogenization (5 min, 3 s on, 5 s off, 50% amp) on ice. Clear cell lysate was collected after centrifugation at 1,400g for 5 min. Protein concentration was measured by BCA assay, and then the cell lysate was diluted to 2 μg μl⁻¹ with lysis buffer. A sample of 15 μl (containing 30 μg of protein) was loaded into each well of a 96-well plate, with BAX spike-in samples receiving 1 μM of recombinant BAX protein. To each of three sets of BAX spike-in K562 lysate, 5 μl of vehicle (DMSO) or CBI1 solution in lysis buffer was added to a final concentration of either 25 μM or 50 μM (0.5% DMSO) and incubated for 1 h at room temperature. To all nine samples, 5 μl of desthiobiotin iodoacetamide (DBIA) solution in lysis buffer was added to a concentration of 500 μM and incubated in the dark for 1 h at room temperature. Subsequently, 3 μl of SP3 beads (1:1 mixture of hydrophobic and hydrophilic type, 50 mg ml⁻¹, cat. no. 45152105050250 and cat. no. 65152105050250, Cytiva), and then 30 μl of ~98% ethanol supplemented with 20 mM DTT was added. The lysate–bead mixture was incubated for 10 min with mild shaking before placing the plate on a magnetic stand to aspirate the supernatant. The beads were washed once with 80% ethanol and resuspended in 30 μl of lysis buffer supplemented with 20 mM iodoacetic acid (IAA) and incubated in the dark for 30 min with vigorous shaking. Then, 60 μl of ~98% ethanol supplemented with 20 mM DTT was added to the mixture, followed by mild shaking and two washes of 80% ethanol. The aqueous phase was then removed, and 30 μl of 200 mM EPPS buffer (pH 8.5) containing 0.3 μg of Lys-C was added. After 3-h incubation at 37 °C, 5 μl of EPPS buffer containing 0.3 μg of trypsin was added and incubated with beads at 37 °C overnight. To the mixture of digested peptides and beads, 9 μl of acetonitrile and 6 μl of TMTpro 18-plex reagent (cat. no. A52045, Thermo Fisher Scientific) at 10 μg μl⁻¹ were sequentially added, followed by gentle mixing for 60 min at room temperature. The reaction was quenched by adding 7 μl of 5% hydroxylamine. The aqueous phase containing TMT-labeled peptides was combined and dried using a SpeedVac. The resulting sample was desalted using a 100-mg Sep-Pak column. Desalted TMT-labeled peptides were resuspended in 460 μl of 100 mM HEPES buffer (pH 7.4) and incubated with 80 μl of Pierce High Capacity Streptavidin Agarose (cat. no. 20359) at room temperature for 3 h. The mixture was then loaded onto an Ultrafree-MC centrifugal filter (hydrophilic PTFE, 0.22-μm pore size) and centrifuged at 1,000g for 30 s. The beads were sequentially washed twice with 300 μl of 100 mM HEPES (pH 7.4) containing 0.05% NP-40, three times with 350 μl of 100 mM HEPES (pH 7.4) and once with 400 μl of water. Peptides were sequentially eluted by incubation with (1) elution buffer (80% acetonitrile, 0.1% formic acid) for 20 min at room temperature; (2) elution buffer for 10 min at room temperature; and (3) elution buffer for 10 min at 72 °C. The combined eluate was dried in a SpeedVac and desalted using StageTips before liquid chromatography–high-field asymmetric waveform ion mobility spectrometry–tandem mass spectrometry (LC–FAIMS–MS/MS) analysis.

LC–FAIMS–MS/MS analysis.

Samples were resuspended in LC–MS loading buffer (5% acetonitrile and 5% formic acid) and loaded onto a 100-μm capillary column packed with 30 cm of Accucore 150 resin (2.6 μm, 150 Å, Thermo Fisher Scientific). Enriched cysteine samples were separated using a 180-min method on a Proxeon NanoLC-1200 UPLC system. Cysteine data were collected using a high-resolution MS2 method on an Orbitrap Eclipse mass spectrometer coupled with a FAIMS Pro device. Data were collected alternating between a set of three FAIMS compensation voltages (CVs). For the single-shot analysis starting with 10 μg of lysate per TMT channel, only one set of CV values (−60 V, −45 V and −35 V) was used. For the double-shot analysis starting with 20 μg of lysate per TMT channel, two sets of CV values (−60 V, −45 V and −35 V; −70 V, −55 V and −30 V) were used. MS1 scans were collected in the Orbitrap with a resolution setting of 60,000, a mass range of 400–1,600 m/z, an AGC at 100% and a maximum injection time of 50 ms. MS2 scans were acquired in Top Speed mode with a cycle time of 1 s. Peptide precursors were selected and fragmented using higher-energy collisional dissociation (HCD) with a collision energy of 36. MS2 scans were collected in the Orbitrap with a resolution of 50,000, a fixed scan range of 110–2,000 m/z and a 500% AGC with a maximum injection time of 86 ms. Dynamic exclusion was set to 120 s with a mass tolerance of ±10 p.p.m. The flowthrough was separated using a 60-min method and analyzed by FAIMS–MS/MS using CV set of −80 V, −60 V and −40 V. Data-dependent analyses were performed in the same setting as analyzing cysteine samples except with a dynamic exclusion time of 90 s.

Data analysis for BAX cysteine identification, localization and quantification.

Raw files of cysteine profiling were searched using the open-source Comet search engine (version 2019.01.5)³⁸ with the UniProt human proteome database (downloaded 24 November 2021) with contaminants and reverse decoy sequences appended. Precursor error tolerance was 50 p.p.m., and fragment error tolerance was 0.9 Da. Static modifications include Cys carboxyamidomethylation (+57.0215) and TMTpro18 (+304.2071) on Lys side chains and peptide N-termini. Methionine oxidation (+15.9949) and DBIA modification on cysteine residues (+239.1628) were allowed as variable modifications. Peptide spectral matches were filtered to a peptide false discovery rate (FDR) of less than 1% using linear discriminant analysis employing a target-decoy strategy^39,40. Resulting peptides were further filtered to obtain a 1% protein FDR at the entire dataset level (including all plexes in an experiment)⁴¹. Cysteine-modified peptides were filtered for site localization using the AScorePro algorithm with a cutoff of 13 (P < 0.05) as previously described^42,43. Overlapping peptide sequences generated from different charge states, retention times and tryptic termini were grouped together into a single entry. A single quantitative value was reported, and only unique peptides were reported. Reporter ion intensities were adjusted to correct for impurities during synthesis of different TMT reagents according to the manufacturer’s specifications. For quantification of each MS3 spectrum, a total sum signal-to-noise of all reporter ions of 180 (TMTPro 18-plex) was required. Peptide quantitative values were normalized so that the sum of the signal for all proteins in each channel was equal to account for sample loading differences (column normalization).

Molecular dynamics simulations

Molecular dynamics simulations were initiated from a solution structure of human BAX (Protein Data Bank (PDB) ID: 1F16). Protein dynamics were described using the CHARMM36 protein^44-47 force field, and the CBI1 ligand was modeled using parameters from CHARMM36 and CGenFF⁴⁸. Simulation assemblies for both wild-type and F116A BAX were constructed with and without CBI1 covalently bound to C126 via a disulfide bond. Systems were solvated usinga TIP3P water model^49-51 (15-Å padding) and supplemented with 0.15 M K⁺/Cl⁻ to ensure electrical neutrality^52,53. The initial simulation cells measured 76 × 77 × 79 ų in volume. Simulations were driven by NAMD2.14 code⁵⁴, which uses a fully periodic BBK-type integrator with velocity rescaling^55,56. We employed rigid bond constraints^57,58 to achieve numerical stability at a timestep of δt = 2 fs (ref. 59). Our NVT calculations applied a Langevin thermostat to heavy atoms for temperature control (T = 300.0 K; damping γ = 1.0 ps⁻¹)⁵⁶ while NPT simulations controlled pressure with a Langevin piston (target P = 101.325 kPa; period = 100.0 fs; decay time = 50.0 fs)⁶⁰. Anisotropic cell fluctuations were allowed, and multiple time-stepping was used for non-bonded interactions⁶¹, with short-range interactions evaluated every 2 fs and full electrostatics every 4 fs. Short-range interactions were cut off at 1.2 nm and smoothed with sigmoidal rescaling for atoms separated by more than 1.0 nm. Smoothed particle meshEwald (PME) was used for long-range electrostatics (80 × 80 × 80 grid)^62-64. Equilibration began with a 2,000-step conjugate gradient (CG) minimization, followed by 1.0 ns of NVT equilibration with all protein atoms fixed. Next, a harmonic constraint potential (k_prot = 5 kcal mol⁻¹ Å⁻²) was applied to hold protein atoms near their initial positions, and 2,000 steps of CG minimization and 2.0 ns of NPT equilibration were used to relax the protein. Finally, all constraints were removed, and production calculations were executed for 300 ns in the NPT ensemble. Structural samples were captured every 5 ps, and the first 100 ns of the trajectories were discounted as an initial equilibration period to provide 200 ns of production simulation.

The difference distance map ${\Delta }_{ij}$ between protein conformers with distances matrices $C$ and $D$ was calculated as ${\Delta }_{ij}=C-D$, where the entry $C_{ij}=N^{-1}{\sum }_{n=1}^N{d}_{ij}(t_n)$ is the average of the distance $d_{ij}(t_n)$ between residues ($i$ and $j$, taken over $N$ trajectory samples at times $\{t_n\}$. Differential distance maps were calculated using custom, Python-based analysis tools, where the NumPy⁶⁵ library performs the numerical analysis, and Matplotlib⁶⁶ is used for data visualization. The contact probability between CBI1 and a given residue of BAX is calculated as the percentage of trajectory samples for which (at least) one atom of CBI1 and one atom of a protein residue have a separation d ≤ 5 Å (refs. 65,66).

HMQC NMR

Uniformly ¹⁵N-labeled recombinant wild-type BAX and BAX C126A were generated as previously described^2,3. In brief, BL21 (DE3) E. coli cells were grown in LB-carbenicillin media to OD of 0.6–0.8, pelleted by centrifugation and resuspended in M9 minimal media (3.2 g L⁻¹ KH₂PO₄, 12.8 g L⁻¹ Na₂HPO₄-7H₂O, 0.5 g L⁻¹ NaCl, 0.4% glucose, 1× trace minerals (Teknova), 2 mM MgSO₄, 100 μM CaCl₂) containing 100 mg L⁻¹ carbenicillin and 1 g L^{−1 15}NH₄Cl (Cambridge Isotope Laboratories). Resuspended cultures were shaken at 37 °C for 30 min, and then 1.0 mM IPTG was added to induce ¹⁵N-BAX expression at 30 °C for 4 h. ¹⁵N-labeled protein was purified as described above for recombinant BAX. Protein samples (20 μM, 2% DMSO; 40 μM, 4% DMSO) were subjected to NMR analyses in the presenceor absence of the indicated molar ratio of CBI1 or N-CBI1 in 25 mM sodium phosphate, 50 mM NaCl, pH 6.2, 10% D₂O. Correlation ¹H-¹⁵N HMQC spectra^67,68 were acquired at 25 °C on a Bruker Avance II 600 MHz NMR spectrometer equipped with a cryogenic probe, processed in Topspin (Bruker) and analyzed using CcpNmr Analysis version 3 (ref. 69). Chemical shift perturbations were calculated in ppm by applying the formula $\sqrt{0.5\ast \left({(\Delta H)}^2+{(\Delta N∕5)}^2\right)}$ where $\Delta \mathrm{H}$ and $\Delta \mathrm{N}$ are the respective changes in ppm of ¹H or ¹⁵N for the indicated cross-peak⁷⁰. The absence of a bar indicates no chemical shift difference or the presence of a proline or residue that is either overlapped or not assigned. BAX cross-peaks were assigned based on previous reports^2,15. A beige bar indicates a cross-peak that experienced severe signal attenuation or very large chemical shift perturbation upon small-molecule incubation. The significance thresholds for the chemical shift changes were calculated based on the average chemical shift across all residues plus 1 s.d. and 2 s.d., in accordance with standard methods⁷¹.

Purification of BAX/BAK-deficient mouse liver mitochondria

Liver mitochondria from Alb-cre^posBax^f/fBak^−/− mice were isolated as described previously²². In brief, collected livers were sliced, washed with isolation buffer (250 μM sucrose, 0.1 mM EGTA, 10 mM Tris, pH 7.4) and dounce homogenized, and the resultant suspension was centrifuged for 10 min at 800g and 4 °C to remove residual fatty tissue. Mitochondria were then pelleted by centrifugation for 10 min at 7,000g and 4 °C. The pellet was resuspended, rinsed in cold isolation buffer and centrifuged for 10 min at 7,000g and 4 °C. The resultant pellet was resuspended in 1 ml of isolation buffer, and total protein concentration was measured by BCA protein assay (Pierce). After quantification, the mitochondrial suspension was diluted 25-fold in preservation buffer (300 mM trehalose, 10 mM KCl, 1 mM EGTA, 1 mM EDTA, 0.1% BSA, 10 mM HEPES, pH 7.7) and centrifuged for 20 min at 2,433g and 4 °C. The mitochondrial pellet was then resuspended in 25 ml of preservation buffer and centrifuged for 20 min at 2,433g and 4 °C. Finally, the mitochondrial pellet was resuspended in preservation buffer at a final protein concentration of 50 mg ml⁻¹, flash frozen on dry ice and stored at −80 °C until use.

Mitochondrial cytochrome $\mathit{c}$ release assay

Frozen BAX/BAK-deficient mitochondria were thawed on ice and then buffer exchanged from preservation buffer to mitochondrial assay buffer (200 mM mannitol, 68 mM sucrose, 1 mM EDTA, 110 mM KCl, 10 mM HEPES, pH 7.4). Recombinant wild-type or F116A BAX (final concentration of BAX proteins as indicated in the figures) was pre-incubated with vehicle (0.08–0.64% DMSO) or inhibitors (CBI1, BAI1 or EO; final concentrations as indicated in the figures) for 1 h at room temperature, added to the mitochondria in the presence or absence of tBID (0.02 eq) and then incubated for 45 min (final concentration of mitochondria = 0.25 mg ml⁻¹). Control wells contained mitochondria and buffer, with and without added 0.5% Triton X-100 (v/v). After incubation, plates were centrifuged for 15 min at 2,068g, and 50 μl of supernatant was analyzed using the Rat/Mouse Cytochrome $c$ Quantikine ELISA Kit (R&D Systems) according to the manufacturer’s instructions using SoftMax Pro 7.0.2 software. Fraction cytochrome $c$ released into the supernatant (fraction cytochrome $c$ release) was calculated according to the following equation: fraction cytochrome $c\text{release}=(\text{cyto}{c}_{\text{sup}}-\text{cyto}{c}_{\text{buffer}})∕(\text{cyto}{c}_{\max }-\text{cyto}{c}_{\text{buffer}})$, where $\text{cyto}{c}_{\text{sup}}-\text{cyto}{c}_{\text{buffer}}$ and $\text{cyto}{c}_{\max }-\text{cyto}{c}_{\text{buffer}}$ represent the amount of cytochrome $c$ specifically released into the supernatant upon treatment with the indicated conditions or 0.5% (v/v) Triton X-100, respectively. The above animal-derived materials (liver mitochondria) were obtained from Alb-cre^posBax^f/fBak^−/− mice in accordance with the guidelines and regulations set forth by the Institutional Animal Care and Use Committee of the Dana-Farber Cancer Institute and in compliance with approved study protocol no. 06-004. Animals were housed in microisolater cages at an ambient temperature of 68–79 °F and humidity of 30–70%, with 12-h light/dark cycles.

HDX MS

HDX MS analyses of BAX proteins were performed essentially as described previously¹⁶, with details specific to the experiments provided per the recommended format⁷² (Source Data Fig. 5 and Supplementary Data 1). BAX aliquots were freshly prepared in biological replicates before HDX MS analysis. Wild-type or F116A BAX (5 μM) was incubated with vehicle (1.25% DMSO) or compound (125 μM) for 1 h at 23 °C (room temperature) in liposomal release assay buffer (20 mM HEPES-KOH, 150 mM KCl, 5 mM MgCl₂, pH 7.2). Deuterium labeling of 2 μl of the equilibration reaction was initiated with an 18-fold dilution into D₂O labeling buffer (36 μl, 20 mM HEPES-KOH, 150 mM KCl, 1 mM MgCl₂, pD 7.6, 99.9% D₂O). After each labeling time (10 s, 1 min and 10 min) at 23 °C, the labeling reaction was quenched with the addition of 38 μl of ice-cold quenching buffer (0.8 M guanidinium chloride, 0.8% (v/v) formic acid, pH 2.0, water) and analyzed immediately using a Waters HDX system coupled to a Waters Synapt G2-Si HDMS^E mass spectrometer in ion mobility mode. Deuterated and control samples were digested online at 15 °C using an AffiPro Nepenthesin-2 column, and peptides were trapped and desalted on a VanGuard Pre-Column trap for 3 min at 100 μl min⁻¹. Peptides were then eluted from the trap using a 5–35% gradient of acetonitrile with 0.1% formic acid over 6 min at a flow rate of 100 μl min⁻¹ and separated using an ACQUITY UPLC HSS T3, 1.8 μm, 1.0 mm × 50-mm column. The main cooling chamber of the HDX system was held at 0.0 ± 0.1 °C for the entire time of the measurements. Peptides were identified from replicate HDMS^E analyses (as detailed in Source Data Fig. 5 and Supplementary Data 1) of undeuterated control samples using PLGS 3.0.1 (Waters Corporation). Peptide masses were identified from searches using non-specific cleavage of a custom database containing the sequence of BAX (UniProt ID: Q07812), no missed cleavages, no PTMs, a low energy threshold of 135, an elevated energy threshold of 35 and an intensity threshold of 500, with no set minimum peptide length. No FDR control was performed. The peptides identified in PLGS (excluding all neutral loss and in-source fragmentation identifications) were then filtered in DynamX 3.0 (Waters Corporation) implementing a minimum products per amino acid cutoff of 0.25 and at least one consecutive product ion (Source Data Fig. 5 and Supplementary Data 1). Those peptides meeting the filtering criteria to this point were further processed by DynamX 3.0 (Waters Corporation). The relative amount of deuterium in each peptide was determined using DynamX 3.0 by subtracting the centroid mass of the undeuterated form of each peptide from the deuterated form, at each timepoint, for each condition. These relative deuterium uptake values are shown in Source Data Fig. 5 and Supplementary Data 1 and were used to generate the corresponding uptake graphs and difference maps. The error of determining the average deuterium incorporation for each peptide was at or below ±0.25 Da. Deuterium levels were not corrected for back exchange and, thus, are reported as relative⁷³.

In vitro mitochondrial translocation assay

Frozen BAX/BAK-deficient mouse liver mitochondria were prepared, thawed and buffer exchanged as described above. Recombinant BAX proteins were pre-incubated with vehicle (DMSO) or CBI1 for 1 h at room temperature and then added to mitochondria in the presence or absence of tBID. The reactions were incubated for 45 min at room temperature (final concentrations: BAX = 2 μM, tBID =40 nM, mitochondria = 4 mg ml⁻¹, CBI1 as indicated in the figures). The samples were then centrifuged at 8,000g for 15 min, and the supernatant was collected. The mitochondrial pellets were resuspended in an equal volume of mitochondrial assay buffer supplemented with 1% (v/v) Triton X-100 to solubilize the mitochondrial fraction. Supernatant and mitochondrial fractions were then analyzed by SDS-PAGE and western blot using 2D2 mouse monoclonal BAX antibody (Santa Cruz Biotechnology, sc-20067; RRID: AB_626726; 1:200). Western analysis using B-6 mouse monoclonal VDAC1 antibody (Santa Cruz Biotechnology, sc-390996; RRID: AB_2750920; 1:200) was also performed to verify the mitochondrial fraction. Densitometric analyses of BAX western blots were performed using ImageJ to quantify the fraction of BAX in the mitochondrial pellet for each sample, calculated as ${\mathrm{B}}_{\mathrm{p}}∕({\mathrm{B}}_{\mathrm{s}}+{\mathrm{B}}_{\mathrm{p}})$, where ${\mathrm{B}}_{\mathrm{p}}$ is the intensity of BAX in the pellet minus the background intensity of the blot in the corresponding lane, and ${\mathrm{B}}_{\mathrm{s}}$ is the intensity of BAX in the supernatant minus the background intensity of the blot in the corresponding lane.

Competitive inhibition of t-2-hex lipidation of BAX by CBI1

Recombinant wild-type or F116A BAX (5 μM) was incubated with vehicle (1.25% ethanol) or t-2-hex (2.5 mM, Cayman Chemical) in the presence of CBI1 or N-CBI1 at the indicated doses or vehicle (1% DMSO) in BAX FPLC buffer for 2 h at 37 °C. For time-dependent experiments, CBI1 was added 5 min, 30 min or 60 min before the addition of t-2-hex. Cy5 hydrazide (5 μM, Kerafast) was then added, and the solution was incubated for 1 h at room temperature. The reaction was quenched by adding Tris base (219 mM), followed by treatment with NaCNBH₃ (6 mM) to stabilize the hydrazide adduct. The reactions were then analyzed by SDS-PAGE, and Cy5 fluorescence (excitation 635 nm, emission 665 nm, 100-μm pixel size) was detected using a Typhoon FLA 9500 (GE Healthcare Life Sciences). Total monomeric BAX was then determined by Coomassie stain.

Statistical methods

GraphPad Prism and Microsoft Excel software were used for data analysis and for calculating mean, s.d. and s.e.m. values.

Biological materials

Plasmids are available upon request to the corresponding author.

Extended Data

Extended Data Fig. 1 ∣. Comparative inhibitory effects of small molecule hits from the disulfide tethering screen on tBID-triggered BAX-mediated liposomal release.

a-d, Liposomal release in response to BAX WT, tBID, the indicated molecule, or the combination of tBID and BAX WT pre-incubated with or without increasing doses of 1C18 (a), 1A18 (b), 1E4 (c), or 3A20 (d). Data are mean ± s.e.m. for experiments performed in technical quadruplicate and repeated using independent preparations of liposomes, proteins, and molecules with similar results.

Extended Data Fig. 2 ∣. Effect of CBI1 on BIM SAHB-triggered BAX-mediated liposomal release.

Dose-responsive inhibition of BIM SAHB-triggered, BAX-mediated liposomal poration (dark gray) upon addition of CBI1 (blue). Data are mean ± s.e.m. for experiments performed in technical quadruplicate and repeated using independent preparations of liposomes, protein, and compounds with similar results.

Extended Data Fig. 3 ∣. No inhibitory effect of CBI1 on Fos-12-induced oligomerization of BAX.

SEC profiles of monomeric BAX (gray) and Fos-12-induced BAX oligomer in the presence (blue) or absence (black) of CBI1. The experiment was performed twice with independent preparations of protein, detergent, and small molecule with similar results.

Extended Data Fig. 4 ∣. NMR analysis of the BAX/CBI1 interaction.

a, Measured chemical shift changes of ¹⁵N-BAX (20 μM) upon addition of CBI1 (5:1 of CBI1:BAX), plotted as a function of BAX residue number. Chemical shift changes above the 2 s.d. cutoff (significance threshold of 0.0536 p.p.m.) are colored maroon and those above the 1 s.d. cutoff (significance threshold of 0.0347 p.p.m.) are colored red. Residues whose cross peaks experienced prominent signal attenuation or chemical shift perturbation upon CBI1 incubation are colored beige. b-c, Chemical shift perturbations of BAX cross peaks corresponding to residues F116 (b, dashed box) and S118 (c) upon addition of CBI1. Whereas BAX exhibited one cross peak each for F116 and S118 (gray), the addition of CBI1 at a CBI1:BAX ratio of 5:1 resulted in the appearance of a second cross peak for each residue (red). Upon incubation with CBI1 at a CBI1:BAX ratio of 10:1, the original cross peaks shifted completely to the second locations (blue). In contrast, D142 (b) experienced little to no change in its corresponding cross peak upon addition of CBI1. d, Prominent signal attenuation or chemical shift perturbation of A81 in wild-type BAX upon CBI1 titration, as reflected by disappearance of the cross peak. e, In contrast, in the context of BAX C126A, the A81 cross peak demonstrates progressive migration upon CBI1 titration, consistent with fast exchange between the unbound and bound forms of BAX C126A, as expected for non-covalent interaction.

Extended Data Fig. 5 ∣. Influence of CBI1 on the conformational dynamics of the BAX α1-α2 loop.

a, A difference distance matrix plot derived from molecular dynamics simulations of wild-type BAX in the presence or absence of CBI1 demonstrated the greatest effect of small molecule C126-derivatization on the protein dynamics of the α1-α2 loop. b-c, Representative images from the molecular dynamics (MD) simulations of BAX in the absence (b) and presence (c) of CBI1, demonstrating distinct positioning of the α1-α2 loop (purple).

Extended Data Fig. 6 ∣. Comparative NMR analyses of the BAX WT/CBI1, BAX WT/N-CBI1, and BAX C126A/CBI1 interactions.

a-c, Measured chemical shift changes of ¹⁵N-BAX WT (40 μM) upon addition of CBI1 (5:1 small molecule:protein) (a), ¹⁵N-BAX WT (40 μM) upon addition of N-CBI1 (5:1 small molecule:protein) (b), and ¹⁵N-BAX C126A (40 μM) upon addition of CBI1 (5:1 small molecule:protein) (c), plotted as a function of BAX residue number. Chemical shift changes above the 2 s.d. cutoff (significance thresholds of 0.0499, 0.0187, 0.01579 p.p.m. for a, b, c, respectively) are colored maroon and those above the 1 s.d. cutoff (significance thresholds of 0.0329, 0.0125, 0.0107 p.p.m. for a, b, c, respectively) are colored red. Residues whose cross peaks experienced prominent signal attenuation or chemical shift perturbation upon small molecule incubation are colored beige.

Extended Data Fig. 7 ∣. Relative impact of covalent vs. non-covalent small molecule interaction on BAX F116A-mediated liposomal permeabilization.

a-b, Comparative inhibitory effects of CBI1 and N-CBI1 on liposomal release by BAX F116A (a) or BAX F116A/C126A (b). Data are mean ± s.e.m. for experiments performed in technical quadruplicate and repeated using independent preparations of liposomes, proteins, and small molecules with similar results.

Extended Data Fig. 8 ∣. CBI1 reverses the conformational activation of BAX F116A.

a-b, Difference distance matrix plots derived from molecular dynamics simulations of BAX F116A compared to BAX WT (a) and BAX F116A in the presence or absence of CBI1 (b) demonstrated striking reversal of the auto-activating conformational changes induced by BAX F116A mutagenesis upon CBI1 covalent derivatization of C126.

Extended Data Fig. 9 ∣. CBI1 blocks the mitochondrial translocation of BAX F116A.

a-b, Distribution of BAX F116A(a) or BAX F116A/C126A (b) (2 μM) between supernatant and BAX/BAK-deficient mitochondrial fractions, as detected by BAX western analysis after pre-treating BAX proteins with escalating doses of CBI1 (200 nM-4 μM, lanes 2-7), incubation with mitochondria, isolation of the supernatant and pellet fractions by centrifugation, and SDS PAGE. Isolation of the mitochondrial pellet fraction was verified by VDAC1 western analysis. The experiment was performed three times using independent preparations of mitochondria, proteins, and small molecule.

Extended Data Fig. 10 ∣. CBI1 blocks t-2-hex lipidation and induced homo-oligomerization of BAX F116A.

Incubation of BAX F116A (5 μM) with t-2-hex (2.5 mM) in the presence or absence of increasing amounts (5 μM-100 μM) of CBI1 (lanes 3-7) or N-CBI1 (lanes 9-13) for 2 hours at 37 °C followed by detection of lipidated BAX by addition of Cy5-hydrazide, gel electrophoresis, and fluorescence scan. BAX protein lipidation, t-2-hex-induced homo-oligomerization (as reflected by laddering), and comparative dose-responsive suppression by CBI1 and N-CBI1, was detected by fluorescence scan of the indicated BAX mixtures (top and middle panels). The influence of t-2-hex and co-treatment with CBI1 or N-CBI1 on the level of monomeric BAX F116A was monitored by protein stain of the electrophoresed BAX mixtures (bottom panel). The experiment was performed twice using independent preparations of protein, lipid, and small molecules.

Data availability

All data generated or analyzed for this study are included in the manuscript and its Supplementary Information. HDX MS data have been deposited to the ProteomeXchange Consortium via the PRIDE⁷⁴ partner repository with dataset identifier PXD040917 and are also included in the manuscript as Source Data Fig. 5. The NMR structure of full-length BAX, corresponding to PDB ID: 1F16, was used in this study. Source data are provided with this paper.

Code availability

No code was generated for this study.